Integrating machine learning and symbolic regression for predicting damage initiation in hybrid FRP bolted connections.

The increasing adoption of machine learning (ML) in fiber-reinforced polymer (FRP) composite design has led to a reliance on black-box models, which achieve high predictive accuracy but lack interpretability. Python symbolic regression (PySR) offers a solution by deriving explicit equations that reveal the governing mechanics of composite structures. This study focuses on hybrid FRP bolted connections, which are rapidly adopted in the industry but remain insufficiently addressed in academic research. To address this gap, a framework was developed to identify key design parameters and predict damage initiation loads by integrating experimental testing, finite element modeling (FEM), and ML. Feature selection and ML models analyzed the dataset, providing insights that guided PySR in deriving interpretable equations. Hybrid L-joint specimens were fabricated and tested to determine damage initiation loads, with results validating FEM models in ABAQUS. A design of experiments approach structured the dataset, and feature selection identified key factors influencing joint performance. ML models assessed dataset quality, with Huber regression emerging as the best-performing model. Based on insights from feature analysis and ML models, PySR derived a compact, interpretable equation that provided greater accuracy and deeper physical insights than the Huber model. This equation aids hybrid L-joint design by improving the understanding of damage initiation mechanics. Beyond predictive accuracy, the findings highlight the model's scalability to different bolt sizes, equally spaced row of bolts, and stacking sequences. This study demonstrates the potential of interpretable ML in structural engineering applications, particularly for hybrid composite-metal joints, where transparent models are essential for design optimization and predictive accuracy.